SCRATCH is a web server for predicting protein tertiary structure and structural features. It offers predictions for secondary structure, solvent accessibility, disordered regions, domains, disulfide bridges, mutation stability, residue contacts, and tertiary structure. Users submit an amino acid sequence and select desired predictions, and results are emailed to them. The server is available at http://www.igb.uci.edu/servers/psss.html. The SCRATCH suite combines machine learning methods, evolutionary information, fragment libraries from the PDB, and energy functions to predict structural features and tertiary structures. It includes modules for secondary structure, solvent accessibility, contacts, domains, disordered regions, mutation stability, disulfide bridges, contact maps, and tertiary structure prediction. The predictors use supervised learning with curated datasets from the PDB. They employ 1D and 2D RNNs, SVMs, and graph matching algorithms. DOMpro predicts domain boundaries in three steps, using statistical methods and domain boundaries from the CATH database. CMAPpro predicts contact maps using variant architectures and thresholds. For tertiary structure prediction, 3Dpro combines predicted structural features, fragment libraries, and energy terms from the PDB. It uses simulated annealing and crankshaft moves to search conformational space. The best model is selected based on energy scores. New features in SCRATCH include improved predictors for domains, disordered regions, mutation stability, and disulfide bonds. The server handles up to 175,000 jobs since 2000, with performance evaluated on various test sets. The server processes input via an HTML form, with results emailed or as attachments. Output includes predictions for structural features, contact maps, and tertiary structures. The server is implemented with Perl scripts and a Linux cluster, using PSI-BLAST and BLAST for sequence alignment. SCRATCH is suitable for sequences without high sequence homology to known structures. It performs well on targets without structural templates. The server is continuously updated with new methods and improved predictors. The server handles sequences up to 400 amino acids for contact maps and tertiary structure prediction. It supports high-throughput projects and longer sequences through offline runs. Future work includes adding a CM component and improving tertiary structure prediction methods. Funding comes from various grants and awards.SCRATCH is a web server for predicting protein tertiary structure and structural features. It offers predictions for secondary structure, solvent accessibility, disordered regions, domains, disulfide bridges, mutation stability, residue contacts, and tertiary structure. Users submit an amino acid sequence and select desired predictions, and results are emailed to them. The server is available at http://www.igb.uci.edu/servers/psss.html. The SCRATCH suite combines machine learning methods, evolutionary information, fragment libraries from the PDB, and energy functions to predict structural features and tertiary structures. It includes modules for secondary structure, solvent accessibility, contacts, domains, disordered regions, mutation stability, disulfide bridges, contact maps, and tertiary structure prediction. The predictors use supervised learning with curated datasets from the PDB. They employ 1D and 2D RNNs, SVMs, and graph matching algorithms. DOMpro predicts domain boundaries in three steps, using statistical methods and domain boundaries from the CATH database. CMAPpro predicts contact maps using variant architectures and thresholds. For tertiary structure prediction, 3Dpro combines predicted structural features, fragment libraries, and energy terms from the PDB. It uses simulated annealing and crankshaft moves to search conformational space. The best model is selected based on energy scores. New features in SCRATCH include improved predictors for domains, disordered regions, mutation stability, and disulfide bonds. The server handles up to 175,000 jobs since 2000, with performance evaluated on various test sets. The server processes input via an HTML form, with results emailed or as attachments. Output includes predictions for structural features, contact maps, and tertiary structures. The server is implemented with Perl scripts and a Linux cluster, using PSI-BLAST and BLAST for sequence alignment. SCRATCH is suitable for sequences without high sequence homology to known structures. It performs well on targets without structural templates. The server is continuously updated with new methods and improved predictors. The server handles sequences up to 400 amino acids for contact maps and tertiary structure prediction. It supports high-throughput projects and longer sequences through offline runs. Future work includes adding a CM component and improving tertiary structure prediction methods. Funding comes from various grants and awards.